NFL Data Analytics & Visualization¶

image.png

READING DATA¶

In [1]:
#importing libraries

import gc
import os
import pandas as pd
import numpy as np
import datetime as dt

import matplotlib.pyplot as plt
import matplotlib.cm as cm
import seaborn as sns

import plotly.graph_objects as go
from plotly.subplots import make_subplots
import plotly.express as px
import plotly.offline

from colorama import Fore, Style, init
from pprint import pprint

import warnings
warnings.filterwarnings('ignore')
In [2]:
# Color printing
# inspired by https://www.kaggle.com/code/ravi20076/sleepstate-eda-baseline
def PrintColor(text:str, color = Fore.BLUE, style = Style.BRIGHT):
    "Prints color outputs using colorama using a text F-string";
    print(style + color + text + Style.RESET_ALL);
    
# inspired by https://www.kaggle.com/code/rishabh15virgo/cmi-dss-first-impression-data-understanding-eda
def summarize_dataframe(df):
    summary_df = pd.DataFrame(df.dtypes, columns=['dtypes'])
    summary_df['missing#'] = df.isna().sum().values*100
    summary_df['missing%'] = (df.isna().sum().values*100)/len(df)
    summary_df['uniques'] = df.nunique().values
    summary_df['first_value'] = df.iloc[0].values
    summary_df['last_value'] = df.iloc[len(df)-1].values
    summary_df['count'] = df.count().values
    #sum['skew'] = df.skew().values
    desc = pd.DataFrame(df.describe().T)
    summary_df['min'] = desc['min']
    summary_df['max'] = desc['max']
    summary_df['mean'] = desc['mean']
    return summary_df
In [3]:
import os
for dirname, _, filenames in os.walk('/kaggle/input'):
    for filename in filenames:
        PrintColor(os.path.join(dirname, filename))

DESCRIBING DATA¶

This chapter conveys a summary of each data set in the 2024 Data Bowl, a list of key variables to join on, and a description of each variable. The tracking data is provided by the NFL Next Gen Stats team. The pff_missedTackles column in the tackles data is provided by Pro Football Focus. The player tracking in the dataset includes data from Weeks 1-9 of the 2022 NFL season. Data will show the location, speed, and acceleration of all 22 players on the field, along with football location. Additional PFF scouting data and NFL advanced stats such as expected points and win probability are also included.

File descriptions

Game data: The games.csv contains the teams playing in each game. The key variable is gameId.

Play data: The plays.csv file contains play-level information from each game. The key variables are gameId and playId.

Player data: The players.csv file contains player-level information from players that participated in any of the tracking data files. The key variable is nflId.

Tackles data: The tackles.csv file contains player-level tackle information for each game and play. The key variables are gameId, playId, and nflId.

Tracking data: Files tracking_week.csv contain player tracking data from week number. The key variables are gameId, playId, and nflId.

Initial Data Inspection¶

Games Data Inspection

In [4]:
games = pd.read_csv("C:/Users/vinit/Downloads/VIRTI/games.csv")
games.head(5).style.set_caption("Sample of the games data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[4]:
Sample of the games data
  gameId season week gameDate gameTimeEastern homeTeamAbbr visitorTeamAbbr homeFinalScore visitorFinalScore
0 2022090800 2022 1 09/08/2022 20:20:00 LA BUF 10 31
1 2022091100 2022 1 09/11/2022 13:00:00 ATL NO 26 27
2 2022091101 2022 1 09/11/2022 13:00:00 CAR CLE 24 26
3 2022091102 2022 1 09/11/2022 13:00:00 CHI SF 19 10
4 2022091103 2022 1 09/11/2022 13:00:00 CIN PIT 20 23
In [5]:
games.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 136 entries, 0 to 135
Data columns (total 9 columns):
 #   Column             Non-Null Count  Dtype 
---  ------             --------------  ----- 
 0   gameId             136 non-null    int64 
 1   season             136 non-null    int64 
 2   week               136 non-null    int64 
 3   gameDate           136 non-null    object
 4   gameTimeEastern    136 non-null    object
 5   homeTeamAbbr       136 non-null    object
 6   visitorTeamAbbr    136 non-null    object
 7   homeFinalScore     136 non-null    int64 
 8   visitorFinalScore  136 non-null    int64 
dtypes: int64(5), object(4)
memory usage: 9.7+ KB
In [6]:
summarize_dataframe(games).style.background_gradient(cmap='Purples')
Out[6]:
  dtypes missing# missing% uniques first_value last_value count min max mean
gameId int64 0 0.000000 136 2022090800 2022110700 136 2022090800.000000 2022110700.000000 2022098922.117647
season int64 0 0.000000 1 2022 2022 136 2022.000000 2022.000000 2022.000000
week int64 0 0.000000 9 1 9 136 1.000000 9.000000 4.845588
gameDate object 0 0.000000 27 09/08/2022 11/07/2022 136 nan nan nan
gameTimeEastern object 0 0.000000 8 20:20:00 20:15:00 136 nan nan nan
homeTeamAbbr object 0 0.000000 32 LA NO 136 nan nan nan
visitorTeamAbbr object 0 0.000000 32 BUF BAL 136 nan nan nan
homeFinalScore int64 0 0.000000 38 10 13 136 3.000000 49.000000 22.669118
visitorFinalScore int64 0 0.000000 35 31 27 136 0.000000 48.000000 20.948529

The charts show data from 136 games played between September 8, 2022, and November 7, 2022 (spanning 9 weeks of Season 2022). The first 8 characters of the 'gameId' attribute indicate the game date in YYYYMMDD format.

On average, teams scored 22.67 points in home games and 20.95 points in away games. The scores for home games ranged from 3 to 49, while the scores for away games ranged from 0 to 48.

Players Data Inspection

In [7]:
players_path = "C:/Users/vinit/Downloads/VIRTI/players.csv"

players = pd.read_csv(players_path)
players.head(5).style.set_caption("Sample of the players data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[7]:
Sample of the players data
  nflId height weight birthDate collegeName position displayName
0 25511 6-4 225 1977-08-03 Michigan QB Tom Brady
1 29550 6-4 328 1982-01-22 Arkansas T Jason Peters
2 29851 6-2 225 1983-12-02 California QB Aaron Rodgers
3 30842 6-6 267 1984-05-19 UCLA TE Marcedes Lewis
4 33084 6-4 217 1985-05-17 Boston College QB Matt Ryan
In [8]:
players.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1683 entries, 0 to 1682
Data columns (total 7 columns):
 #   Column       Non-Null Count  Dtype 
---  ------       --------------  ----- 
 0   nflId        1683 non-null   int64 
 1   height       1683 non-null   object
 2   weight       1683 non-null   int64 
 3   birthDate    1204 non-null   object
 4   collegeName  1683 non-null   object
 5   position     1683 non-null   object
 6   displayName  1683 non-null   object
dtypes: int64(2), object(5)
memory usage: 92.2+ KB
In [9]:
summarize_dataframe(players).style.background_gradient(cmap='Purples')
Out[9]:
  dtypes missing# missing% uniques first_value last_value count min max mean
nflId int64 0 0.000000 1683 25511 55241 1683 25511.000000 55241.000000 48221.702317
height object 0 0.000000 16 6-4 6-2 1683 nan nan nan
weight int64 0 0.000000 179 225 280 1683 153.000000 380.000000 245.724302
birthDate object 47900 28.461081 985 1977-08-03 nan 1204 nan nan nan
collegeName object 0 0.000000 226 Michigan Coastal Carolina 1683 nan nan nan
position object 0 0.000000 19 QB DT 1683 nan nan nan
displayName object 0 0.000000 1672 Tom Brady C.J. Brewer 1683 nan nan nan

We discovered that:

  • The dataset has details about 1683 players who took part in the games (with birth dates specified for 1204 players).
  • There are 1672 different names for the players, suggesting that 11 players share names with others in the dataset.
  • There are 985 unique birth dates recorded in the dataset.
  • The players come from 226 different colleges.

Plays Data Inspection

In [10]:
plays_path = "C:/Users/vinit/Downloads/VIRTI/plays.csv"

plays = pd.read_csv(plays_path)
plays.head(5).style.set_caption("Sample of the plays data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[10]:
Sample of the plays data
  gameId playId ballCarrierId ballCarrierDisplayName playDescription quarter down yardsToGo possessionTeam defensiveTeam yardlineSide yardlineNumber gameClock preSnapHomeScore preSnapVisitorScore passResult passLength penaltyYards prePenaltyPlayResult playResult playNullifiedByPenalty absoluteYardlineNumber offenseFormation defendersInTheBox passProbability preSnapHomeTeamWinProbability preSnapVisitorTeamWinProbability homeTeamWinProbabilityAdded visitorTeamWinProbilityAdded expectedPoints expectedPointsAdded foulName1 foulName2 foulNFLId1 foulNFLId2
0 2022100908 3537 48723 Parker Hesse (7:52) (Shotgun) M.Mariota pass short middle to P.Hesse to 50 for 9 yards (K.Neal). 4 1 10 ATL TB ATL 41 7:52 21 7 C 6.000000 nan 9 9 N 69 SHOTGUN 7.000000 0.747284 0.976785 0.023215 -0.006110 0.006110 2.360609 0.981955 nan nan nan nan
1 2022091103 3126 52457 Chase Claypool (7:38) (Shotgun) C.Claypool right end to PIT 37 for 3 yards (C.Awuzie). 4 1 10 PIT CIN PIT 34 7:38 14 20 nan nan nan 3 3 N 76 SHOTGUN 7.000000 0.416454 0.160485 0.839515 -0.010865 0.010865 1.733344 -0.263424 nan nan nan nan
2 2022091111 1148 42547 Darren Waller (8:57) D.Carr pass short middle to D.Waller to LV 45 for 15 yards (N.Adderley). 2 2 5 LV LAC LV 30 8:57 10 3 C 11.000000 nan 15 15 N 40 I_FORM 6.000000 0.267933 0.756661 0.243339 -0.037409 0.037409 1.312855 1.133666 nan nan nan nan
3 2022100212 2007 46461 Mike Boone (13:12) M.Boone left tackle to DEN 44 for 7 yards (J.Abram; D.Deablo). 3 2 10 DEN LV DEN 37 13:12 19 16 nan nan nan 7 7 N 47 SINGLEBACK 6.000000 0.592704 0.620552 0.379448 -0.002451 0.002451 1.641006 -0.043580 nan nan nan nan
4 2022091900 1372 47857 Devin Singletary (8:33) D.Singletary right guard to TEN 32 for 3 yards (N.Jones; O.Adeniyi). TEN-O.Adeniyi was injured during the play. His return is Questionable. O.Adeniyi walks off. 2 1 10 BUF TEN TEN 35 8:33 7 7 nan nan nan 3 3 N 75 I_FORM 7.000000 0.470508 0.836290 0.163710 0.001053 -0.001053 3.686428 -0.167903 nan nan nan nan
In [11]:
plays.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 12486 entries, 0 to 12485
Data columns (total 35 columns):
 #   Column                            Non-Null Count  Dtype  
---  ------                            --------------  -----  
 0   gameId                            12486 non-null  int64  
 1   playId                            12486 non-null  int64  
 2   ballCarrierId                     12486 non-null  int64  
 3   ballCarrierDisplayName            12486 non-null  object 
 4   playDescription                   12486 non-null  object 
 5   quarter                           12486 non-null  int64  
 6   down                              12486 non-null  int64  
 7   yardsToGo                         12486 non-null  int64  
 8   possessionTeam                    12486 non-null  object 
 9   defensiveTeam                     12486 non-null  object 
 10  yardlineSide                      12319 non-null  object 
 11  yardlineNumber                    12486 non-null  int64  
 12  gameClock                         12486 non-null  object 
 13  preSnapHomeScore                  12486 non-null  int64  
 14  preSnapVisitorScore               12486 non-null  int64  
 15  passResult                        6105 non-null   object 
 16  passLength                        5634 non-null   float64
 17  penaltyYards                      615 non-null    float64
 18  prePenaltyPlayResult              12486 non-null  int64  
 19  playResult                        12486 non-null  int64  
 20  playNullifiedByPenalty            12486 non-null  object 
 21  absoluteYardlineNumber            12486 non-null  int64  
 22  offenseFormation                  12482 non-null  object 
 23  defendersInTheBox                 12481 non-null  float64
 24  passProbability                   12149 non-null  float64
 25  preSnapHomeTeamWinProbability     12486 non-null  float64
 26  preSnapVisitorTeamWinProbability  12486 non-null  float64
 27  homeTeamWinProbabilityAdded       12486 non-null  float64
 28  visitorTeamWinProbilityAdded      12486 non-null  float64
 29  expectedPoints                    12486 non-null  float64
 30  expectedPointsAdded               12485 non-null  float64
 31  foulName1                         592 non-null    object 
 32  foulName2                         25 non-null     object 
 33  foulNFLId1                        592 non-null    float64
 34  foulNFLId2                        25 non-null     float64
dtypes: float64(12), int64(12), object(11)
memory usage: 3.3+ MB
In [12]:
summarize_dataframe(plays).style.background_gradient(cmap='Purples')
Out[12]:
  dtypes missing# missing% uniques first_value last_value count min max mean
gameId int64 0 0.000000 136 2022100908 2022102308 12486 2022090800.000000 2022110700.000000 2022098953.855598
playId int64 0 0.000000 3974 3537 905 12486 54.000000 5096.000000 1986.603476
ballCarrierId int64 0 0.000000 480 48723 54572 12486 25511.000000 55158.000000 48072.271664
ballCarrierDisplayName object 0 0.000000 480 Parker Hesse Dameon Pierce 12486 nan nan nan
playDescription object 0 0.000000 12486 (7:52) (Shotgun) M.Mariota pass short middle to P.Hesse to 50 for 9 yards (K.Neal). (12:51) C.Heck reported in as eligible. D.Pierce up the middle to LV 14 for 3 yards (D.Perryman; D.Deablo). 12486 nan nan nan
quarter int64 0 0.000000 5 4 2 12486 1.000000 5.000000 2.550136
down int64 0 0.000000 4 1 3 12486 1.000000 4.000000 1.727054
yardsToGo int64 0 0.000000 32 10 1 12486 1.000000 38.000000 8.469085
possessionTeam object 0 0.000000 32 ATL HOU 12486 nan nan nan
defensiveTeam object 0 0.000000 32 TB LV 12486 nan nan nan
yardlineSide object 16700 1.337498 32 ATL LV 12319 nan nan nan
yardlineNumber int64 0 0.000000 50 41 17 12486 1.000000 50.000000 29.573122
gameClock object 0 0.000000 898 7:52 12:51 12486 nan nan nan
preSnapHomeScore int64 0 0.000000 42 21 3 12486 0.000000 49.000000 11.170671
preSnapVisitorScore int64 0 0.000000 36 7 0 12486 0.000000 48.000000 9.862967
passResult object 638100 51.105238 2 C nan 6105 nan nan nan
passLength float64 685200 54.877463 65 6.000000 nan 5634 -9.000000 61.000000 5.455982
penaltyYards float64 1187100 95.074483 23 nan nan 615 -15.000000 15.000000 -2.091057
prePenaltyPlayResult int64 0 0.000000 86 9 3 12486 -26.000000 98.000000 7.582012
playResult int64 0 0.000000 97 9 3 12486 -61.000000 98.000000 7.218565
playNullifiedByPenalty object 0 0.000000 2 N N 12486 nan nan nan
absoluteYardlineNumber int64 0 0.000000 99 69 27 12486 11.000000 109.000000 60.426077
offenseFormation object 400 0.032036 7 SHOTGUN I_FORM 12482 nan nan nan
defendersInTheBox float64 500 0.040045 11 7.000000 8.000000 12481 1.000000 11.000000 6.424085
passProbability float64 33700 2.699023 11554 0.747284 0.101552 12149 0.005607 0.996206 0.603845
preSnapHomeTeamWinProbability float64 0 0.000000 12429 0.976785 0.726639 12486 0.001049 0.999175 0.556236
preSnapVisitorTeamWinProbability float64 0 0.000000 12429 0.023215 0.273361 12486 0.000825 0.998951 0.443764
homeTeamWinProbabilityAdded float64 0 0.000000 12410 -0.006110 -0.015028 12486 -0.521181 0.471527 0.000110
visitorTeamWinProbilityAdded float64 0 0.000000 12410 0.006110 0.015028 12486 -0.471527 0.521181 -0.000110
expectedPoints float64 0 0.000000 12104 2.360609 4.685974 12486 -2.366872 6.558894 2.258360
expectedPointsAdded float64 100 0.008009 12450 0.981955 0.251580 12485 -9.986150 8.698986 0.297367
foulName1 object 1189400 95.258690 25 nan nan 592 nan nan nan
foulName2 object 1246100 99.799776 10 nan nan 25 nan nan nan
foulNFLId1 float64 1189400 95.258690 420 nan nan 592 33107.000000 55157.000000 47759.856419
foulNFLId2 float64 1246100 99.799776 25 nan nan 25 43586.000000 54650.000000 50000.560000

We discovered that there are details for 12,486 plays in the dataset, and each play has its own distinct and not empty description.

Tackles Data Inspection

In [13]:
tackles_path = "C:/Users/vinit/Downloads/VIRTI/tackles.csv"

tackles = pd.read_csv(tackles_path)
tackles.head(5).style.set_caption("Sample of the tackles data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[13]:
Sample of the tackles data
  gameId playId nflId tackle assist forcedFumble pff_missedTackle
0 2022090800 101 42816 1 0 0 0
1 2022090800 393 46232 1 0 0 0
2 2022090800 486 40166 1 0 0 0
3 2022090800 646 47939 1 0 0 0
4 2022090800 818 40107 1 0 0 0
In [14]:
tackles.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 17426 entries, 0 to 17425
Data columns (total 7 columns):
 #   Column            Non-Null Count  Dtype
---  ------            --------------  -----
 0   gameId            17426 non-null  int64
 1   playId            17426 non-null  int64
 2   nflId             17426 non-null  int64
 3   tackle            17426 non-null  int64
 4   assist            17426 non-null  int64
 5   forcedFumble      17426 non-null  int64
 6   pff_missedTackle  17426 non-null  int64
dtypes: int64(7)
memory usage: 953.1 KB
In [15]:
summarize_dataframe(tackles).style.background_gradient(cmap='Purples')
Out[15]:
  dtypes missing# missing% uniques first_value last_value count min max mean
gameId int64 0 0.000000 136 2022090800 2022091901 17426 2022090800.000000 2022110700.000000 2022098971.441123
playId int64 0 0.000000 3943 101 3845 17426 54.000000 5096.000000 1982.974578
nflId int64 0 0.000000 800 42816 46199 17426 33131.000000 55241.000000 47602.719442
tackle int64 0 0.000000 2 1 0 17426 0.000000 1.000000 0.569207
assist int64 0 0.000000 2 0 0 17426 0.000000 1.000000 0.315276
forcedFumble int64 0 0.000000 2 0 0 17426 0.000000 1.000000 0.005681
pff_missedTackle int64 0 0.000000 2 0 1 17426 0.000000 1.000000 0.119936

We discovered details on 17,426 tackles and attempted tackles in the dataset. Out of the 1,683 players mentioned, 800 were part of these events. In about 56.92% of the tackle records, a player actually made a tackle, and in 31.53% of the records, a player assisted in a tackle.

Insights from Games Data¶

Games by Dates

In [16]:
games_count = games['gameDate'].value_counts().reset_index()
games_count.columns = ['date', 'games']
games_count = games_count.sort_values('date')
plt.figure(figsize=(6,6))
sns.barplot(data=games_count, y='date', x='games', palette='viridis')
plt.title('Number of games by dates')
plt.xlabel('Number of games')
plt.ylabel('Date')
plt.show()

  • Most games in the dataset happen on Sundays, like on September 11, 2022, September 18, 2022, and so on.

  • The largest number of games on a single day, totaling 14, occurred on specific dates: September 11, September 25, October 2, and October 9, 2022.

Games by Weeks

In [17]:
check = games['week'].value_counts().reset_index()
check.columns = ['week', 'games']
check = check.sort_values('week')
fig = px.line(  check, x='week',y='games', markers=True, title='Number of games over weeks',
labels={'week': 'Week', 'games': 'Number of Games'})
fig.update_layout( width=600, height=600)
fig.show()
24681313.51414.51515.516
Number of games over weeksWeekNumber of Games
plotly-logomark

We find that

  • the biggest number of the games per week (16) occured on weeks 1-5
  • the smallest number of of the games per week (13) was observed on the week 9

Home and Away Games by Teams

In [18]:
dfg_home = games['homeTeamAbbr'].value_counts().reset_index().rename(columns={'index': 'team', 'homeTeamAbbr': 'number_of_home_games'})
dfg_away = games['visitorTeamAbbr'].value_counts().reset_index().rename(columns={'index': 'team', 'visitorTeamAbbr': 'number_of_away_games'})
outer_merged = dfg_home.merge(dfg_away, on='team', how='outer')
merged_melted = outer_merged.melt(id_vars='team', var_name='game_type', value_name='number_of_games')
fig = px.bar(merged_melted, x='team', y='number_of_games', color='game_type',
             labels={'number_of_games': 'Number of Games', 'game_type': 'Game Type'},
             title='Home vs. Visiting Games by Teams',
             color_discrete_map={'number_of_home_games': px.colors.qualitative.Prism[0],
                                 'number_of_away_games': px.colors.qualitative.Prism[1]},height=400, width=800,barmode='group')  

fig.update_layout(showlegend=True,margin=dict(t=30, b=0, l=5, r=5),template="plotly_white",)
fig.show()
NOLAWASJAXCLEATLDALARITBNYJDETCARMINMIANEINDPHIGBDENNYGKCCHICINBALLACSEAHOUTENSFLVBUFPIT0123456
Game Typenumber_of_home_gamesnumber_of_away_gamesHome vs. Visiting Games by TeamsteamNumber of Games
plotly-logomark

Home Games

In [19]:
agg_dict_1 = {
    "min_home_score": pd.NamedAgg(column='homeFinalScore', aggfunc='min'),
    "max_home_score": pd.NamedAgg(column='homeFinalScore', aggfunc='max'),
    "avg_home_score": pd.NamedAgg(column='homeFinalScore', aggfunc=np.mean),

}

agg_data = games.groupby("homeTeamAbbr").agg(**agg_dict_1).reset_index()
agg_data.head()
Out[19]:
homeTeamAbbr min_home_score max_home_score avg_home_score
0 ARI 12 42 22.600000
1 ATL 17 37 26.200000
2 BAL 19 38 25.000000
3 BUF 27 41 35.333333
4 CAR 15 24 19.600000

  1. Buffalo Bills Dominance: BUF leads with an impressive home score of 35.33, showcasing dominance.

  2. Detroit Lions Variability: DET shows a wide range (15 to 45) in home scores, indicating variability.

  3. New England Patriots Consistency: NE maintains consistent home performance with a narrow score range.

  4. Las Vegas Raiders Strength: LV excels with a high max home score (38) and a solid average of 31.

  5. Minnesota Vikings Excellence: MIN demonstrates excellence at home with an average score of 28.5.

  6. Kansas City Chiefs Balance: KC displays balanced home performance, suggesting stability.

  7. San Francisco 49ers Consistency: SF shows consistent home performances with a narrow score range.

  8. Arizona Cardinals Competitiveness: ARI is competitive with a respectable average home score of 22.6.

  9. Tampa Bay Buccaneers Variation: TB has varied home performances, indicating a mix of strengths and challenges.

  10. League-Wide Diversity: The data showcases diverse home performances, highlighting unique team strengths.

Away Game

In [20]:
agg_dict_2 = {
    "min_away_score": pd.NamedAgg(column='visitorFinalScore', aggfunc='min'),
    "max_away_score": pd.NamedAgg(column='visitorFinalScore', aggfunc='max'),
    "avg_away_score": pd.NamedAgg(column='visitorFinalScore', aggfunc=np.mean)
}

agg_data = games.groupby("visitorTeamAbbr").agg(**agg_dict_2).reset_index()
agg_data.head()
Out[20]:
visitorTeamAbbr min_away_score max_away_score avg_away_score
0 ARI 9 29 22.50
1 ATL 15 27 21.50
2 BAL 20 37 27.00
3 BUF 17 31 22.80
4 CAR 10 34 20.25

  1. Kansas City Chiefs Dominance: KC stands out with an impressive average away score of 36.5, showcasing road dominance.

  2. Buffalo Bills Consistency: BUF maintains strong away performance, with an average score of 22.8, highlighting consistency.

  3. San Francisco 49ers Versatility: SF displays adaptability with a solid average away score of 20.4, showcasing versatility.

  4. New Orleans Saints Balance: NO strikes a balance with a well-rounded average away score of 25, indicating stability.

  5. Los Angeles Chargers Strength: LAC excels on the road, boasting an impressive average away score of 27, showcasing strength.

  6. Seattle Seahawks High-Scoring: SEA stands out for high-scoring away games, with an average of 31, highlighting offensive prowess.

  7. Jacksonville Jaguars Resilience: JAX shows road resilience, with an average away score of 27, indicating the ability to compete.

  8. New England Patriots Consistency: NE maintains consistency on the road, with an average away score of 21.6.

  9. Tampa Bay Buccaneers Variation: TB displays varied away performances, with an average score of 15, suggesting a mix of challenges and successes.

  10. Green Bay Packers Efficiency: GB exhibits road efficiency with a decent average away score of 13.6.

Insights from Player Data¶

In [21]:
players.head().style.set_caption("Sample of the Player data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[21]:
Sample of the Player data
  nflId height weight birthDate collegeName position displayName
0 25511 6-4 225 1977-08-03 Michigan QB Tom Brady
1 29550 6-4 328 1982-01-22 Arkansas T Jason Peters
2 29851 6-2 225 1983-12-02 California QB Aaron Rodgers
3 30842 6-6 267 1984-05-19 UCLA TE Marcedes Lewis
4 33084 6-4 217 1985-05-17 Boston College QB Matt Ryan

Player Data Transformation and Feature Engineering Before delving into the data analytics, let's do a quick player data preprocessing and feature engineering:

In [22]:
# cast height to the metric system - namely, in m

new_players = players['height'].str.split('-',expand=True)
new_players.columns=['Foot','Inch']
new_players['Foot'] = new_players['Foot'].astype(int)
new_players['Inch'] = new_players['Inch'].astype(int)
new_players['Foot'] = new_players['Foot'] * 30.48
new_players['Inch'] = new_players['Inch'] * 2.54
new_players['height_cm'] = new_players['Foot'] + new_players['Inch']


players['height'] = new_players['height_cm'] / 100
# cast the weights to metric units - namely, to kg
players['weight'] = players['weight'] * 0.45



formats = ["%Y-%m-%d", "%m/%d/%Y"]
parsed_dt = pd.to_datetime(players.birthDate, format=formats[0], errors='coerce')
for format in formats[1:]:
    parsed_dt = parsed_dt.fillna(pd.to_datetime(players.birthDate, format=format, errors='coerce'))

players['birthDate'] = parsed_dt

def group_position(position):
    if position in ['QB', 'T', 'TE', 'WR', 'C', 'G', 'RB', 'FB']:
        return 'offensive'
    return 'defensive'


players['position_group'] = players['position'].apply(group_position)

players.height = players.height.round(2)
players.weight = players.weight.round(2)

# formula inherited from https://www.calculatorsoup.com/calculators/health/bmi-calculator.php
players['mass_index'] = players['weight']/( players['height'] * players['height'] )

players.mass_index = players.mass_index.round(2)

def group_index(index):
    bmi_lower_fence = 21.64
    bmi_q1 = 27.18
    bmi_median = 29.87
    bmi_q3 = 34.53
    
    if index < bmi_q1:
        return 'S'
    elif bmi_q1 <= index < bmi_median:
        return 'M'
    elif bmi_median <= index < bmi_q3:
        return 'L'
    else:
        return 'XL'

players['bmi_group'] = players['mass_index'].apply(group_index)

today = dt.datetime.today()

players['age'] = players['birthDate'].apply(
               lambda x: today.year - x.year - 
               ((today.month, today.day) < (x.month, x.day)) 
               )

players.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1683 entries, 0 to 1682
Data columns (total 11 columns):
 #   Column          Non-Null Count  Dtype         
---  ------          --------------  -----         
 0   nflId           1683 non-null   int64         
 1   height          1683 non-null   float64       
 2   weight          1683 non-null   float64       
 3   birthDate       1204 non-null   datetime64[ns]
 4   collegeName     1683 non-null   object        
 5   position        1683 non-null   object        
 6   displayName     1683 non-null   object        
 7   position_group  1683 non-null   object        
 8   mass_index      1683 non-null   float64       
 9   bmi_group       1683 non-null   object        
 10  age             1204 non-null   float64       
dtypes: datetime64[ns](1), float64(4), int64(1), object(5)
memory usage: 144.8+ KB
In [23]:
del new_players
# Returns the number of
# objects it has collected
# and deallocated
collected = gc.collect()
 
# Prints Garbage collector
# as 0 object
print("Garbage collector: collected",
          "%d objects." % collected)# Importing gc module

Garbage collector: collected 1150 objects.
In [24]:
players.head().style.set_caption("Sample of the Player data (after transformation and FE)"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[24]:
Sample of the Player data (after transformation and FE)
  nflId height weight birthDate collegeName position displayName position_group mass_index bmi_group age
0 25511 1.930000 101.250000 1977-08-03 00:00:00 Michigan QB Tom Brady offensive 27.180000 M 46.000000
1 29550 1.930000 147.600000 1982-01-22 00:00:00 Arkansas T Jason Peters offensive 39.630000 XL 41.000000
2 29851 1.880000 101.250000 1983-12-02 00:00:00 California QB Aaron Rodgers offensive 28.650000 M 40.000000
3 30842 1.980000 120.150000 1984-05-19 00:00:00 UCLA TE Marcedes Lewis offensive 30.650000 L 39.000000
4 33084 1.930000 97.650000 1985-05-17 00:00:00 Boston College QB Matt Ryan offensive 26.220000 S 38.000000

Players by Age

In [25]:
fig = px.box(players, x="age",
                   title='Summary Statistics on Players by Age',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
2530354045
Summary Statistics on Players by Ageage
plotly-logomark
  • 75% of players in the dataset are aged 30 or younger.
  • The oldest player in the dataset is 46 years old.
  • The youngest players mentioned are 23 years old.
  • Players aged 37 to 46 are considered 'outliers' in terms of age distribution.

Age of players grouped by defensive/offensive positions

In [26]:
fig = px.box(players, x="position_group", y="age",
                   title='Summary Statistics on Players by Age and Offensive/Defensive Position',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
offensivedefensive2530354045
Summary Statistics on Players by Age and Offensive/Defensive Positionposition_groupage
plotly-logomark

There is no notable distinction in age distribution between players in offensive and defensive positions. Additionally, it is noteworthy that offensive positions exhibit a higher proportion of older players, as indicated by the 'outliers' in the diagram above.

Average age of players in different colleges.

In [27]:
avg_age  = players.groupby(['collegeName'])['age'].mean()
frame = {'collegeName': avg_age.index,
    'avg_player_age': avg_age}
dfg_agg = pd.DataFrame(frame)
dfg_agg = dfg_agg.sort_values('avg_player_age', ascending=True)
dfg_agg
Out[27]:
collegeName avg_player_age
collegeName
St. John's, Minn. St. John's, Minn. 25.0
Norfolk State Norfolk State 25.0
Wagner Wagner 25.0
Berry Berry 25.0
Missouri S&amp;T Missouri S&amp;T 25.0
... ... ...
Northwest Missouri State Northwest Missouri State NaN
Sam Houston State Sam Houston State NaN
Texas A&amp;M-Commerce Texas A&amp;M-Commerce NaN
Tusculum Tusculum NaN
West Florida West Florida NaN

226 rows × 2 columns

In [28]:
import plotly.express as px

fig = px.bar(
    dfg_agg, 
    y='collegeName', 
    x="avg_player_age", 
    orientation='h', 
    title='Average Age of Players in College Teams', 
    height=600, 
    width=600,
    color='avg_player_age',  # Color based on average age
    color_continuous_scale='Viridis',  # Choose a color scale
    labels={'avg_player_age': 'Average Age'},
    template='plotly_white',  # Use a white background template
)
fig.show()
0102030St. John's, Minn.BrownWake ForestMassachusettsWestern IllinoisMemphisHawaiiWestern MichiganValdosta StateSan Diego StateFindlayCharleston, W. Va.Louisiana-LafayetteLouisiana TechSouth CarolinaOld DominionNorthwesternSouthern ArkansasIllinoisKentuckyMichigan StateNorthern IllinoisDelawareMontana StateRutgersCanisiusPresbyterianFramingham StateTusculum
262830323436Average AgeAverage Age of Players in College TeamsAverage AgecollegeName
plotly-logomark

Observations:

  • The top average player ages are found in Troy College (36) and Regina, Can. (34).
  • Seven colleges have the lowest average player age, which is 25.
  • Some colleges, including West Florida, North Carolina at Pembroke, and Florida A&M, lack player birth dates, preventing the calculation of player ages.

Players by BMI¶

In [29]:
fig = px.box(players, x="mass_index",
                   title='Summary Statistics on Players by Body Mass Index (BMI)',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
2530354045
Summary Statistics on Players by Body Mass Index (BMI)mass_index
plotly-logomark

Observations and Insights Regarding Player Health in the NFL:

  • The mass index (BMI) exhibits a roughly binomial distribution.
  • The minimal BMI is recorded at 21.64, indicating a lower limit.
  • The median BMI value is 29.87, providing a central point of reference.
  • 75% of players in the dataset have a BMI equal to or below 34.53, suggesting a majority within a certain range.
  • The upper limit (fence) for BMI is set at 44.56.
  • Notable outliers are present, with three players having exceptionally high BMIs of 45.62, 45.69, and 47.24. These outliers could indicate unique health considerations or anomalies in the dataset.

By integrating these strategies, teams can enhance player well-being and potentially improve overall game performance.

BMI of players grouped by defensive/offensive positions

In [30]:
fig = px.box(players, x="position_group", y="mass_index",
                   title='Summary Statistics on Players by BMI and Offensive/Defensive Position',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
offensivedefensive2530354045
Summary Statistics on Players by BMI and Offensive/Defensive Positionposition_groupmass_index
plotly-logomark

We can see slight difference in the BMI summary statistics for different position groups.

  • Median value of BMI is slightly higher for the players in defensive positions (29.86 vs. 29.92 for offensive positions)
  • Q3 value of BMI is lower for the players in defensive positions (33.03 vs. 35.07 for offensive positions)
  • Upper-fennce value of BMI is higher for the players in offensive positions (43.63 vs. 41.92 for defensive positions)
  • There is a number of 'outlier' players with high mass_index values in defensive positions

Average BMI for players who study in different colleges

In [31]:
avg_bmi  = players.groupby(['collegeName'])['mass_index'].mean()
frame = {'collegeName': avg_bmi.index,
    'avg_player_bmi': avg_bmi}
dfg_agg = pd.DataFrame(frame)
dfg_agg = dfg_agg.sort_values('avg_player_bmi', ascending=True)
fig = px.bar(
    dfg_agg, 
    y='collegeName', 
    x="avg_player_bmi", 
    orientation='h', 
    title='Average BMI of players by the college graduated', 
    height=600, 
    width=600,
    color_discrete_sequence=px.colors.qualitative.Prism,)
fig.show()
010203040TusculumCentral ArkansasGeorgia StateHoly CrossSouth AlabamaCalifornia, Pa.Sacramento StatePittsburghSouth Dakota StatePrincetonMiamiWayne State, Mich.Iowa StateAuburnHoustonNorth Dakota StateFloridaSouthern CaliforniaTexasFlorida StateDelawareClemsonKentuckyTexas-El PasoLouisiana-LafayetteIllinois StateRiceGramblingNorth Carolina at Pembroke
Average BMI of players by the college graduatedavg_player_bmicollegeName
plotly-logomark

We see that

  • the players with the highest average BMI study at Regina, Can. college
  • the players with the lowest average BMI study at Tusculum college

Association between the BMI groups and the player position¶

In [32]:
agg_data = players[["position_group", "bmi_group"]].groupby(["position_group", "bmi_group"]).size().reset_index(name="count")
# define figure element
fig = px.sunburst(
    agg_data,
    values='count',
    path=["position_group", "bmi_group"],
    title="BMI Group-to-Player Position Group Association",
    color="position_group",
    height=600,
    color_discrete_sequence=px.colors.qualitative.Prism,
    #color_continuous_scale=px.colors.sequential.Viridis
)
# display the figure
fig.show()
defensiveoffensiveLSMXLXLMSL
BMI Group-to-Player Position Group Association
plotly-logomark

We infer that

  • There are more XL players (with the quite big value of BMI) and M players (with the medium value of BMI) within offensive positions
  • There are more players with BMI in small and large groups (bmi_group = 'S' and 'L') in defensive positions

Insights from Tackles Data¶

Tackles per Game

In [33]:
tackles.head().style.set_caption("Sample of the Tackle data"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})

tackle_games = tackles.groupby(['gameId'])['tackle'].agg('sum')

frame = {'game_id': tackle_games.index,
         'tackles': tackle_games}

tackle_games = tackles.groupby(['gameId'])['tackle'].agg('sum')

frame = {'game_id': tackle_games.index,
         'tackles': tackle_games}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['game_id'], axis=1)

 
fig = px.box(agg_data, x="tackles",
                   title='Summary Statistics on Tackles per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
50556065707580859095
Summary Statistics on Tackles per Gametackles
plotly-logomark

We see that the distribution of the number of tackles per game is skewed to the right, and the majority of the games had between 65 and 79 tackles per a game.

  • The Q1 value is 67 tackles per game
  • The median value is 74 tackles per game
  • The Q3 value is 78 tackles per game
  • There are no any outliers detected as for the tackles per game (within the set of 136 represented by the competition dataset)

Top 10 games in terms of the number of tackles per game

In [34]:
agg_data.sort_values(['tackles'], 
                   ascending=False)[
                        ['gameId',
                          'tackles',
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[34]:
  gameId tackles
114 2022103005 94
6 2022091105 91
37 2022092504 91
93 2022101700 90
106 2022102311 89
67 2022100902 88
5 2022091104 86
104 2022102309 86
110 2022103001 86
76 2022100911 85

Bottom 10 games in terms of the number of tackles per game.

In [35]:
agg_data.sort_values(['tackles'], 
                   ascending=True)[
                        [
                          'gameId',
                          'tackles',
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[35]:
  gameId tackles
25 2022091808 51
55 2022100206 53
21 2022091804 57
130 2022110606 58
75 2022100910 58
101 2022102306 59
119 2022103010 60
13 2022091112 60
2 2022091101 60
23 2022091806 61

As we can see,

  • The game with gameId = '2022091808' had the smallest number of tackles per game (51)
  • The game with gameId = '2022103005' had the biggest number of tackles per game (94)

Tackles by players

In [36]:
tackle_players = tackles.groupby(['nflId'])['tackle'].agg('sum')

frame = {'nfl_id': tackle_players.index,
         'tackles': tackle_players}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['nfl_id'], axis=1)

fig = px.histogram(agg_data, x="tackles", marginal="rug",
                   title='Distribution of Tackles by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig = px.box(agg_data, x="tackles",
                   title='Summary Statistics on Tackles by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
0102030405060
Summary Statistics on Tackles by Playerstackles
plotly-logomark

We find that

  • the majority of the players referred (75%) have performed 18 or less tackles during the games listed in the dataset
  • we have a number of remarkable outliers indicating players who committed big amount of tackles (41-61); activities of such players should be investigated in more details to look for possible insightful patterns
  • the overall distribution of the tackles by players is skewed to the left

Assists per Game

In [37]:
assist_games = tackles.groupby(['gameId'])['assist'].agg('sum')

frame = {'game_id': assist_games.index,
         'assists': assist_games}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['game_id'], axis=1)

fig = px.histogram(agg_data, x="assists", marginal="rug",
                   title='Distribution of Assists per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig = px.box(agg_data, x="assists",
                   title='Summary Statistics on Assists per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
203040506070
Summary Statistics on Assists per Gameassists
plotly-logomark

We find that

  • The distribution of the assists per game is close to normal
  • The majority of the games has 30-54 assists per game
  • There is a clear outlier in the dataset (a game with the extra big number of assists committed - 71)
  • The Q1 value is 33.5 assists per game
  • The median value is 40 assists per game
  • The Q3 value is 46 assists per game
  • The outlier game (with 71 assists committed) is clearly displayed on the box plot

Top 10 games in terms of the number of assists per a game

In [38]:
agg_data.sort_values(['assists'], 
                   ascending=False)[['gameId','assists',]][:10].style.background_gradient(cmap='seismic')
Out[38]:
  gameId assists
101 2022102306 71
59 2022100210 62
127 2022110603 60
24 2022091807 59
124 2022110600 58
23 2022091806 56
21 2022091804 55
28 2022091811 55
35 2022092502 54
64 2022100600 54

From the table above, it is easy to locate the 'outlier' game with 71 assists - namely, it is the game with gameId = '2022102306'. It would be interesting to look at the 'telemetry' (tracking data) for this game in more details, to see the possible hidden patterns behind assists committed.

Bottom 10 games in terms of the number of assists per game

In [39]:
agg_data.sort_values(['assists'], 
                   ascending=True)[
                        [
                          'gameId',
                          'assists',
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[39]:
  gameId assists
25 2022091808 18
114 2022103005 20
20 2022091803 21
30 2022091900 22
104 2022102309 23
3 2022091102 24
125 2022110601 28
103 2022102308 28
65 2022100900 29
120 2022103011 29

As we can infer, the game with the smallest number of assists is the game with gameId = '2022091808' (only 18 assists committed).

Forced Fumble per Game¶

In [40]:
forced_fumble_games = tackles.groupby(['gameId'])['forcedFumble'].agg('sum')

frame = {'game_id': forced_fumble_games.index,
         'forced_fumbles': forced_fumble_games}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['game_id'], axis=1)

fig = px.histogram(agg_data, x="forced_fumbles", marginal="rug",
                   title='Distribution of Forced Fumbles per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
01234010203040506070
Distribution of Forced Fumbles per Gameforced_fumblescount
plotly-logomark

We find that

  • The majority of the games in the dataset (70 out of 136) did not have any forced fumble events at all
  • The minority rest of the games had 1-4 forced fumble events per game

Top 10 games in terms of the forced fumble events

In [41]:
agg_data.sort_values(['forced_fumbles'], 
                   ascending=False)[
                        [
                          'gameId',
                          'forced_fumbles',
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[41]:
  gameId forced_fumbles
45 2022092512 4
15 2022091200 4
11 2022091110 4
1 2022091100 3
38 2022092505 3
46 2022092513 3
8 2022091107 3
71 2022100906 2
127 2022110603 2
98 2022102303 2

Missed Tackles per Game

In [42]:
missed_tackles_games = tackles.groupby(['gameId'])['pff_missedTackle'].agg('sum')

frame = {'game_id': missed_tackles_games.index,
         'missed_tackles': missed_tackles_games}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['game_id'], axis=1)

fig = px.histogram(agg_data, x="missed_tackles", marginal="rug",
                   title='Distribution of Missed Tackles per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig = px.box(agg_data, x="missed_tackles",
                   title='Summary Statistics on Missed Tackles per Game',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
51015202530
Summary Statistics on Missed Tackles per Gamemissed_tackles
plotly-logomark

We find that

  • The distribution of the missed tackles per game is close to normal
  • There were 19 or less missed tackles in 75% of the games
  • The maximal number of missed tackles per a game was 29
  • The minimal number of missed tackles per a game was 5

Tackles per Player (Groupby)¶

In [43]:
tackle_players = tackles.groupby(['nflId'])['tackle'].agg('sum')

frame = {'nfl_id': tackle_players.index,
         'tackles': tackle_players}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['nfl_id'], axis=1)

fig = px.histogram(agg_data, x="tackles", marginal="rug",
                   title='Distribution of Tackles by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig = px.box(agg_data, x="tackles",
                   title='Summary Statistics on Tackles by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
0204060
Summary Statistics on Tackles by Playerstackles
plotly-logomark

We find that

  • The majority of the players referred (75%) have performed 18 or less tackles during the games listed in the dataset
  • We have a number of remarkable outliers indicating players who committed big amount of tackles (41-61); activities of such players should be investigated in more details to look for possible insightful patterns
  • The overall distribution of the tackles by players is skewed to the left
In [44]:
player_names = players[['nflId', 'displayName', 'position']]
In [45]:
agg_data = pd.merge(
    agg_data,
    player_names,
    how="inner",
    on=['nflId']
)

agg_data.sort_values(['tackles'], 
                   ascending=False)[
                        [
                            'nflId',
                            'displayName',
                            'position',
                            'tackles',
                        ]
][:30].style.background_gradient(cmap='seismic')
Out[45]:
  nflId displayName position tackles
492 52435 Jordyn Brooks ILB 61
334 46269 Foyesade Oluokun ILB 59
337 46304 Zaire Franklin OLB 55
63 41243 C.J. Mosley ILB 55
106 42427 Jordan Hicks ILB 54
606 53489 Pete Werner OLB 51
270 46077 Roquan Smith ILB 50
472 48516 T.J. Edwards ILB 49
414 47872 Bobby Okereke ILB 48
605 53487 Nick Bolton ILB 46
237 44925 Eddie Jackson SS 45
358 46669 Jonathan Owens FS 45
279 46091 Rashaan Evans ILB 45
209 44848 Budda Baker SS 44
425 47913 Drue Tranquill ILB 43
531 52527 Mykal Walker ILB 43
482 49410 Jalen Thompson FS 43
695 54500 Roger McCreary CB 42
97 42388 Eric Kendricks ILB 41
446 47996 Donovan Wilson SS 41
143 43325 Myles Jack ILB 41
538 52546 L'Jarius Sneed CB 40
596 53465 Jevon Holland FS 40
29 38577 Bobby Wagner ILB 40
263 45345 Nicholas Morrow OLB 40
521 52497 Cameron Dantzler CB 39
598 53469 Richie Grant SS 39
568 52852 Myles Hartsfield FS 39
584 53445 Zaven Collins OLB 39
332 46259 DeShon Elliott FS 39

As we can see, the players in the top 30 tackle performers list are distributed by their game positions as follows

  • ILB - 14
  • FS - 5
  • OLB - 4
  • SS - 4
  • CB - 3

Players in the top 10 'tackler' list, in turn, play in ILB and OLB positions solely.

Assists per Player (Groupby)¶

In [46]:
assist_players = tackles.groupby(['nflId'])['assist'].agg('sum')

frame = {'nfl_id': assist_players.index,
         'assists': assist_players}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['nfl_id'], axis=1)

fig = px.histogram(agg_data, x="assists", marginal="rug",
                   title='Distribution of Assists by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig = px.box(agg_data, x="assists",
                   title='Summary Statistics on Assists by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.update_layout( width=600, height=600)
fig.show()
010203040
Summary Statistics on Assists by Playersassists
plotly-logomark

We find that

  • The majority of the players referred (75%) have performed 10 or less assists during the games listed in the dataset
  • We have a number of remarkable outliers indicating players who committed big amount of assists (23-39); activities of such
  • Players should be investigated in more details to look for possible insightful patterns
  • The overall distribution of the assists by players is skewed to the left
In [47]:
agg_data = pd.merge(
    agg_data,
    player_names,
    how="inner",
    on=['nflId']
)

agg_data.sort_values(['assists'], 
                   ascending=False)[
                        [
                            'nflId',
                            'displayName',
                            'position',
                            'assists',
                        ]
][:30].style.background_gradient(cmap='seismic')
Out[47]:
  nflId displayName position assists
279 46091 Rashaan Evans ILB 39
617 53509 Divine Deablo OLB 34
270 46077 Roquan Smith ILB 32
690 54492 Devin Lloyd ILB 32
63 41243 C.J. Mosley ILB 32
439 47956 Cole Holcomb ILB 32
322 46212 Ja'Whaun Bentley ILB 31
407 47855 Germaine Pratt OLB 31
276 46086 Derwin James FS 30
337 46304 Zaire Franklin OLB 28
93 42368 Shaq Thompson ILB 28
380 47788 Devin White ILB 27
165 43404 De'Vondre Campbell ILB 26
143 43325 Myles Jack ILB 26
441 47971 David Long ILB 26
179 43503 Elandon Roberts ILB 26
209 44848 Budda Baker SS 26
334 46269 Foyesade Oluokun ILB 26
531 52527 Mykal Walker ILB 25
278 46088 Leighton Vander Esch OLB 25
472 48516 T.J. Edwards ILB 25
622 53532 Ernest Jones ILB 24
129 42929 Alex Singleton OLB 24
97 42388 Eric Kendricks ILB 24
227 44888 Alex Anzalone ILB 24
584 53445 Zaven Collins OLB 24
492 52435 Jordyn Brooks ILB 23
422 47891 Julian Love SS 23
605 53487 Nick Bolton ILB 23
275 46085 Tremaine Edmunds ILB 23

As we can see, the players in the top 30 'assist performers' list are distributed by their game positions as follows

  • ILB - 21
  • OLB - 6
  • SS - 2
  • FS - 1

In turn, the players in the top 10 list of 'assist performers' play in ILB, OLB, and FS positions.

Forced Fumbles Per Player (Groupby)¶

This is a straightforward metric, but it's a good indicator of a defensive player's ability to change the course of a game. More forced fumbles generally mean a player is good at creating turnover opportunities.

In [48]:
forced_fumble_players = tackles.groupby(['nflId'])['forcedFumble'].agg('sum')

frame = {'nfl_id': forced_fumble_players.index,
         'forced_fumbles': forced_fumble_players}
 
# Creating DataFrame by passing Dictionary
agg_data = pd.DataFrame(frame).reset_index()
agg_data = agg_data.drop(columns=['nfl_id'], axis=1)

fig = px.histogram(agg_data, x="forced_fumbles", marginal="rug",
                   title='Distribution of Forced Fumbles by Players',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
012340100200300400500600700
Distribution of Forced Fumbles by Playersforced_fumblescount
plotly-logomark

We can see that

  • The majority of the players listed in the contest dataset did not commit any forced fumbles
  • We can see 79 players who committed 1 forced fumble
  • 9 players performed more then 1 forced fumbles (and their experience could be investigated in more details down the road)
In [49]:
agg_data = pd.merge(
    agg_data,
    player_names,
    how="inner",
    on=['nflId']
)

agg_data.sort_values(['forced_fumbles'], 
                   ascending=False)[
                        [
                            'nflId',
                            'displayName',
                            'position',
                            'forced_fumbles',
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[49]:
  nflId displayName position forced_fumbles
728 54574 Coby Bryant CB 4
366 46711 Ryan Neal SS 2
687 54487 Quay Walker ILB 2
237 44925 Eddie Jackson SS 2
606 53489 Pete Werner OLB 2
645 53592 Darrick Forrest SS 2
723 54562 Kerby Joseph SS 2
187 43700 Jonathan Jones CB 2
603 53481 Jeremiah Owusu-Koramoah OLB 2
507 52469 Kristian Fulton CB 1

Evaluating Player Tackle Performance¶

Data Preprocessing

In [50]:
agg_dict_1 = {
    "tackles": pd.NamedAgg(column='tackle', aggfunc='sum'),
    "assists": pd.NamedAgg(column='assist', aggfunc='sum'),
    "forced_fumbles": pd.NamedAgg(column='forcedFumble', aggfunc='sum'),
    "missed_tackles": pd.NamedAgg(column='pff_missedTackle', aggfunc='sum'),
}

player_tackle_df = tackles.groupby("nflId").agg(**agg_dict_1).reset_index()

# calculate player tackle efficiency
# inspired by https://www.kaggle.com/code/sasakitetsuya/1st-step-data-summary-and-understanding
player_tackle_df['total_tackles'] = player_tackle_df['tackles'] + player_tackle_df['assists']
player_tackle_df['tackle_efficiency'] = player_tackle_df['total_tackles'] / (player_tackle_df['total_tackles'] + player_tackle_df['missed_tackles'])


player_tackles = pd.merge(
    players,
    player_tackle_df,
    how="inner",
    on=['nflId']
)

player_tackles.head().style.set_caption("Sample of the Player data with tackle stats"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[50]:
Sample of the Player data with tackle stats
  nflId height weight birthDate collegeName position displayName position_group mass_index bmi_group age tackles assists forced_fumbles missed_tackles total_tackles tackle_efficiency
0 33131 2.030000 135.000000 1986-09-01 00:00:00 Miami DE Calais Campbell defensive 32.760000 L 37.000000 8 10 0 4 18 0.818182
1 35449 1.900000 136.800000 1987-05-12 00:00:00 California NT Tyson Alualu defensive 37.890000 XL 36.000000 4 4 0 1 8 0.888889
2 35452 1.880000 119.250000 1988-04-03 00:00:00 Michigan DE Brandon Graham defensive 33.740000 L 35.000000 6 6 0 1 12 0.923077
3 35454 1.960000 123.750000 1989-01-01 00:00:00 South Florida DE Jason Pierre-Paul defensive 32.210000 L 35.000000 3 3 0 1 6 0.857143
4 35459 1.780000 83.250000 1988-04-10 00:00:00 Alabama SS Kareem Jackson defensive 26.280000 S 35.000000 28 23 0 9 51 0.850000
In [51]:
del player_tackle_df
# Returns the number of
# objects it has collected
# and deallocated
collected = gc.collect()
 
# Prints Garbage collector
# as 0 object
print("Garbage collector: collected",
          "%d objects." % collected)# Importing gc module

Garbage collector: collected 9560 objects.

Players-to-Tackles Numeric Variable Correlations

In [52]:
sns.set_theme(style="white")

# make a dataframe with only numberic variables, without 'Outcome'
d = player_tackles[['height', 'weight', 'mass_index', 'age', 'tackle_efficiency',
                    'tackles', 'assists', 'forced_fumbles', 'missed_tackles']]

# Compute the correlation matrix for numeric features
corr = d.corr()

# Generate a mask for the upper triangle
mask = np.triu(np.ones_like(corr, dtype=bool))

# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(11, 9))

# Generate a custom diverging colormap
cmap = sns.diverging_palette(230, 20, as_cmap=True)

# Draw the heatmap with the mask and correct aspect ratio
sns.heatmap(corr, mask=mask, cmap=cmap, vmax=.3, center=0, annot=True,
            square=True, linewidths=.5, cbar_kws={"shrink": .5})
Out[52]:
<Axes: >

As we can see, there are no strong correlations between the player characteristics and tackle-related parameters. However, we can see medium-level correlations as per the list below

  • There is a negative medium-level correlation between the number of tackles and the weight of a player (as well as his BMI) to make a tackle
  • There is a negative medium-level correlation between the number of missed tackles and the weight of a player (as well as his BMI) to make a tackle
  • We can see multiple strong- and medium-level correlations between tackle-related metrics
  • It can impose the current practice of lower-weighted players to be involved in tackling the counter-part offensive players.

Tackles by Player Positions

Let's do a bivariative analysis of the relations between the player positions and player-to-tackle-related numeric variables. Note: We are going to investigate defensive positions only since only players in the defensive roles commit tackles.

In [53]:
fig = make_subplots(rows=5, cols=1, subplot_titles=(
    '<b>Summary Statistics on Players by Tackles and Defensive Roles</b>',
    '<b>Summary Statistics on Players by Assists and Defensive Roles</b>',
    '<b>Summary Statistics on Players by Forced Fumbles and Defensive Roles</b>',
    '<b>Summary Statistics on Players by Missed Tackles and Defensive Roles</b>',
    '<b>Summary Statistics on Players by Tackle Efficiency and Defensive Roles</b>',
))

fig.add_trace(go.Box(x=player_tackles['position'],y=player_tackles['tackles']), row=1, col=1)
fig.add_trace(go.Box(x=player_tackles['position'],y=player_tackles['assists']), row=2, col=1)
fig.add_trace(go.Box(x=player_tackles['position'],y=player_tackles['forced_fumbles']), row=3, col=1)
fig.add_trace(go.Box(x=player_tackles['position'],y=player_tackles['missed_tackles']), row=4, col=1)
fig.add_trace(go.Box(x=player_tackles['position'],y=player_tackles['tackle_efficiency']), row=5, col=1)

# Update visual layout
fig.update_layout(
    showlegend=False,
    width=900,
    height=600,
    autosize=False,
    margin=dict(t=15, b=0, l=5, r=5),
    template="plotly_white",
    colorway=px.colors.qualitative.Prism ,
)

# update font size at the axes
fig.update_coloraxes(colorbar_tickfont_size=10)
# Update font in the titles: Apparently subplot titles are annotations (Subplot font size is hardcoded to 16pt · Issue #985)
fig.update_annotations(font_size=12)
# Reduce opacity
fig.update_traces(opacity=0.75)

fig.show()
DENTSSFSOLBDTCBILBMLBDB0204060DENTSSFSOLBDTCBILBMLBDB010203040DENTSSFSOLBDTCBILBMLBDB01234DENTSSFSOLBDTCBILBMLBDB05101520DENTSSFSOLBDTCBILBMLBDB00.51
Summary Statistics on Players by Tackles and Defensive RolesSummary Statistics on Players by Assists and Defensive RolesSummary Statistics on Players by Forced Fumbles and Defensive RolesSummary Statistics on Players by Missed Tackles and Defensive RolesSummary Statistics on Players by Tackle Efficiency and Defensive Roles
plotly-logomark

We find that:

  • The biggest average number of tackles and assists is committed by players in SS (Strong Safety), ILB (Inside Linebacker), and FS (Free Safety) positions

  • Some of the players in OLB (Outside Linebacker), DT (Defensive Tackle), DE (Defensive End), and CB (Cornerback) positions do quite large number of tackles and assists (they are indicated as outlisers in the chart above)

  • Players in MLB (Middle Linebackeer) position are in the middle of the rank in terms of the number of tackles and assists

  • Players in NT (Nose Tackle) and DB (Defensive Back) positions are in the bottom of the rank in terms of the number of tackles and assists

  • It is interesting to match it to the positions of the players in the top 30 'tackle performers' (see details above). We find that

  • The number of the players in OLB position is exceptionally high, in the top 30 list

  • In turn, the players in FS and SS positions are 'under-represented', in the top 30 list

  • The highest median tackle efficiency is observed for players in MLB (0.9329), FS (0.92), NT (0.91667), ILB (0.91107), and DT (0.9091) positions

  • The lowest median tackle efficiency is observed for players in OLB (0.875) position

Tackles by Player BMI Groups¶

In [54]:
fig = make_subplots(rows=5, cols=1, subplot_titles=(
    '<b>Summary Statistics on Players by Tackles and BMI Group</b>',
    '<b>Summary Statistics on Players by Assists and BMI Group</b>',
    '<b>Summary Statistics on Players by Forced Fumbles and BMI Group</b>',
    '<b>Summary Statistics on Players by Missed Tackles and BMI Group</b>',
    '<b>Summary Statistics on Players by Tackle Efficiency and BMI Group</b>',
))

fig.add_trace(go.Box(x=player_tackles['bmi_group'],y=player_tackles['tackles']), row=1, col=1)
fig.add_trace(go.Box(x=player_tackles['bmi_group'],y=player_tackles['assists']), row=2, col=1)
fig.add_trace(go.Box(x=player_tackles['bmi_group'],y=player_tackles['forced_fumbles']), row=3, col=1)
fig.add_trace(go.Box(x=player_tackles['bmi_group'],y=player_tackles['missed_tackles']), row=4, col=1)
fig.add_trace(go.Box(x=player_tackles['bmi_group'],y=player_tackles['tackle_efficiency']), row=5, col=1)

# Update visual layout
fig.update_layout(
    showlegend=False,
    width=900,
    height=600,
    autosize=False,
    margin=dict(t=15, b=0, l=5, r=5),
    template="plotly_white",
    colorway=px.colors.qualitative.Prism ,
)

# update font size at the axes
fig.update_coloraxes(colorbar_tickfont_size=10)
# Update font in the titles: Apparently subplot titles are annotations (Subplot font size is hardcoded to 16pt · Issue #985)
fig.update_annotations(font_size=12)
# Reduce opacity
fig.update_traces(opacity=0.75)

fig.show()
LXLSM0204060LXLSM010203040LXLSM01234LXLSM05101520LXLSM00.51
Summary Statistics on Players by Tackles and BMI GroupSummary Statistics on Players by Assists and BMI GroupSummary Statistics on Players by Forced Fumbles and BMI GroupSummary Statistics on Players by Missed Tackles and BMI GroupSummary Statistics on Players by Tackle Efficiency and BMI Group
plotly-logomark

We find that:

  • The players with the smallest values of the body mass index ('S' group) commit tackles and assists more often
  • Since missed tackles are measured as a subset of tackles, the same predominance of the players from 'S' group is observed for missed tackles as well
  • Players of the group 'M' take the second place in terms of the number of tackles as well as the first place in terms of the number of assists committed
  • There are notable outliers of playes doing extra big number of tackles in each BMI group
  • Forced fumbles are rare events and are 'outliers' by their nature
  • The median tackle efficiency of the players with the smaller BMI values ('S' and 'M' groups) - 0.88 and 0.89 - is slightly lower then the median tackle efficiency of the players with the bigger BMI values ('L' and 'XL' groups, respectively); the latter groups demonstrate the median tackle performance at 0.90+ level

Average tackle Efficiency

In [55]:
agg_dict_2 = {
    "tackle_efficiency": pd.NamedAgg(column='tackle_efficiency', aggfunc='mean'),
    "total_graduates": pd.NamedAgg(column='tackle_efficiency', aggfunc='count'),
}

college_tackle_efficiency_df = player_tackles.groupby("collegeName").agg(**agg_dict_2).reset_index()

college_tackle_efficiency_df = college_tackle_efficiency_df.sort_values('tackle_efficiency', ascending=True)

fig = px.bar(
    college_tackle_efficiency_df, 
    y='collegeName', 
    x="tackle_efficiency", 
    orientation='h', 
    title='Average tackle_efficiency of players by the college graduated', 
    height=900, 
    width=800,
    color_discrete_sequence=px.colors.qualitative.Prism,
)
fig.show()
00.20.40.60.81North TexasAppalachian StateSam Houston StateNorthern IowaFort Hays StateCharleston, W. Va.TulaneWake ForestTennesseePittsburghGeorgiaSouthern IllinoisIowaIllinoisEastern IllinoisArizonaVirginiaAlabama-BirminghamMissouri S&TLouisiana StateCincinnatiIndiana StateNevadaSoutheast MissouriTexasOhio StateGeorgia SouthernMichiganSamfordAuburnVillanovaDelawareIndianaSouthern MethodistKentuckyUtah StateColorado State-PuebloSouthern MississippiOld DominionPrairie ViewRegina, Can.Southern U.Florida A&M
Average tackle_efficiency of players by the college graduatedtackle_efficiencycollegeName
plotly-logomark

Top 10 colleges by the average tackle efficiency of their graduates

In [56]:
college_tackle_efficiency_df.sort_values(['tackle_efficiency'], 
                   ascending=False)[
                        [
                            'collegeName',
                            'tackle_efficiency',
                            'total_graduates'
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[56]:
  collegeName tackle_efficiency total_graduates
93 North Carolina A&T 1.000000 1
123 South Alabama 1.000000 1
107 Old Dominion 1.000000 1
111 Pennsylvania 1.000000 2
101 Northwest Missouri State 1.000000 1
91 Norfolk State 1.000000 2
113 Prairie View 1.000000 1
8 Army 1.000000 1
17 California, Pa. 1.000000 1
42 Framingham State 1.000000 1

Top 10 colleges by the number of their graduates among the players, factored in the player efficiency

In [57]:
college_tackle_efficiency_df.sort_values(['total_graduates', 'tackle_efficiency'], 
                   ascending=False)[
                        [
                            'collegeName',
                            'tackle_efficiency',
                            'total_graduates'
                        ]
][:10].style.background_gradient(cmap='seismic')
Out[57]:
  collegeName tackle_efficiency total_graduates
0 Alabama 0.888278 29
65 Louisiana State 0.879506 25
104 Ohio State 0.904669 24
76 Michigan 0.908371 17
44 Georgia 0.844284 17
57 Iowa 0.854481 16
36 Florida 0.847639 16
110 Penn State 0.800743 16
103 Notre Dame 0.906731 15
161 Washington 0.892252 15

Based on the charts above, we can conclude that:

  • Although 20+ colleges with the tackle efficiency of the graduates at 1.00, the number of their graduates is extremely low (1-2); it means there is no substancial reason to believe in the special coach 'know-hows' in these colleges
  • In turn, the coaches in Kentucky, Oregon, Auburn, Ohio State, Michigan, Notre Dame, Texas, and UCLA seem to crack the code with growing efficient tackler defenders

Tackling Efficiency and Tackle Factor by Defensive Positions

In [58]:
agg_dict_3 = {
    "avg_tackles_by_pos": pd.NamedAgg(column='total_tackles', aggfunc='mean'),
}

position_tackles_df = player_tackles.groupby("position").agg(**agg_dict_3).reset_index() 

def set_average_tackles_by_pos(position):
    avg_tackles = position_tackles_df.loc[
        position_tackles_df['position'] == position, 'avg_tackles_by_pos'].iloc[0]
    return avg_tackles

player_tackles['avg_tackles_by_pos'] = player_tackles['position'].apply(set_average_tackles_by_pos)

player_tackles['tackle_factor'] = player_tackles['total_tackles']/player_tackles['avg_tackles_by_pos']

player_tackles = player_tackles.drop(columns=['avg_tackles_by_pos'])

player_tackles.head(5).style.set_caption("Sample of the Player data with tackle factor stats"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
fig = px.scatter(player_tackles,
                 x="tackle_factor", y="tackle_efficiency", color='position',
                 hover_data=['nflId',
                            'displayName',],
                   title='Tackling Efficiency and Tackle Factor by Position',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
01234500.20.40.60.81
positionDENTSSFSOLBDTCBILBMLBDBTackling Efficiency and Tackle Factor by Positiontackle_factortackle_efficiency
plotly-logomark

It has been demonstrated that,

  • Tackle factor is a very helpful metric to evaluate the tackling player performance relative to other players who occupy the same defensive position.
  • The successful players in OLB role have the highest values of tackle factor, with players in DE and DT taking the second and third places in terms of the players with the maximal values of tackle factor
  • The coaches in Kentucky, Oregon, Auburn, Ohio State, Michigan, Notre Dame, Texas, and UCLA seem to crack the code with growing efficient tackler defenders.
  • There is a logariphmic functional dependency between tackle factor and tackle efficiency.
In [59]:
player_tackles.sort_values(['tackle_factor'], 
                   ascending=False)[
                        [
                            'nflId',
                            'displayName',
                            'position',
                            'tackle_efficiency',
                            'tackle_factor',
                        ]
][:20].style.background_gradient(cmap='seismic')
Out[59]:
  nflId displayName position tackle_efficiency tackle_factor
337 46304 Zaire Franklin OLB 0.873684 5.080859
617 53509 Divine Deablo OLB 0.912500 4.468707
606 53489 Pete Werner OLB 0.901235 4.468707
584 53445 Zaven Collins OLB 0.828947 3.856556
129 42929 Alex Singleton OLB 0.885714 3.795340
485 52415 Derrick Brown DE 0.954545 3.676220
244 44957 Grover Stewart DT 0.937500 3.630000
278 46088 Leighton Vander Esch OLB 0.951613 3.611695
263 45345 Nicholas Morrow OLB 0.855072 3.611695
420 47889 Maxx Crosby DE 0.833333 3.501162
407 47855 Germaine Pratt OLB 0.890625 3.489265
70 41300 Christian Kirksey OLB 0.900000 3.305619
385 47796 Christian Wilkins DT 0.952381 3.226667
417 47881 Quincy Williams OLB 0.842105 2.938328
300 46146 Sam Hubbard DE 0.868421 2.888459
538 52546 L'Jarius Sneed CB 0.900000 2.880873
405 47848 Zach Allen DE 0.969697 2.800930
148 43335 A'Shawn Robinson DT 0.918919 2.742667
66 41263 Demarcus Lawrence DE 0.861111 2.713400
14 37097 Cameron Jordan DE 0.837838 2.713400

Tackling Efficiency and Tackle Factor by BMI Group

In [60]:
fig = px.scatter(player_tackles,
                 x="tackle_factor", y="tackle_efficiency", color='bmi_group',
                 hover_data=['nflId',
                            'displayName', 'position'],
                   title='Tackling Efficiency and Tackle Factor by BMI Group',
                   color_discrete_sequence=px.colors.qualitative.Prism)
fig.show()
01234500.20.40.60.81
bmi_groupLXLSMTackling Efficiency and Tackle Factor by BMI Grouptackle_factortackle_efficiency
plotly-logomark

It has been observed that,

  • OLB players occupy the most of the slots in top 20 list by tackling factor
  • Tackle efficiency and tackling factor do not lineally correlate
  • The exceptionally performning players fitting into XL and L categories in terms of BMI values have higher tackle factor values; it can be interpreted as the heavier players to be more hard to reach the top-performing metrcs in the tackling (and if the do it, they stand out of the croud of the rest of the heavy players more significantly vs. the players in S and M categories)
  • Divine Deablo, the OLB with small BMI, demostrated exceptional results in terms of the combined values of tackle factor and tackle efficiency among the light-weight players

Tackles by Offensive Formation

In [61]:
agg_dict_1 = {
    "tackles": pd.NamedAgg(column='tackle', aggfunc='sum'),
    "assists": pd.NamedAgg(column='assist', aggfunc='sum'),
    "forced_fumbles": pd.NamedAgg(column='forcedFumble', aggfunc='sum'),
    "missed_tackles": pd.NamedAgg(column='pff_missedTackle', aggfunc='sum'),
}

play_tackle_df = tackles.groupby(["gameId", "playId"]).agg(**agg_dict_1).reset_index()


play_tackles = pd.merge(
    plays,
    play_tackle_df,
    how="outer",
    on=["gameId", "playId"]
)

play_tackles.head().style.set_caption("Sample of the Play data with tackle stats"). \
set_properties(**{'border': '1.3px solid blue',
                          'color': 'grey'})
Out[61]:
Sample of the Play data with tackle stats
  gameId playId ballCarrierId ballCarrierDisplayName playDescription quarter down yardsToGo possessionTeam defensiveTeam yardlineSide yardlineNumber gameClock preSnapHomeScore preSnapVisitorScore passResult passLength penaltyYards prePenaltyPlayResult playResult playNullifiedByPenalty absoluteYardlineNumber offenseFormation defendersInTheBox passProbability preSnapHomeTeamWinProbability preSnapVisitorTeamWinProbability homeTeamWinProbabilityAdded visitorTeamWinProbilityAdded expectedPoints expectedPointsAdded foulName1 foulName2 foulNFLId1 foulNFLId2 tackles assists forced_fumbles missed_tackles
0 2022100908 3537 48723 Parker Hesse (7:52) (Shotgun) M.Mariota pass short middle to P.Hesse to 50 for 9 yards (K.Neal). 4 1 10 ATL TB ATL 41 7:52 21 7 C 6.000000 nan 9 9 N 69 SHOTGUN 7.000000 0.747284 0.976785 0.023215 -0.006110 0.006110 2.360609 0.981955 nan nan nan nan 1.000000 0.000000 0.000000 0.000000
1 2022091103 3126 52457 Chase Claypool (7:38) (Shotgun) C.Claypool right end to PIT 37 for 3 yards (C.Awuzie). 4 1 10 PIT CIN PIT 34 7:38 14 20 nan nan nan 3 3 N 76 SHOTGUN 7.000000 0.416454 0.160485 0.839515 -0.010865 0.010865 1.733344 -0.263424 nan nan nan nan 1.000000 0.000000 0.000000 1.000000
2 2022091111 1148 42547 Darren Waller (8:57) D.Carr pass short middle to D.Waller to LV 45 for 15 yards (N.Adderley). 2 2 5 LV LAC LV 30 8:57 10 3 C 11.000000 nan 15 15 N 40 I_FORM 6.000000 0.267933 0.756661 0.243339 -0.037409 0.037409 1.312855 1.133666 nan nan nan nan 1.000000 0.000000 0.000000 0.000000
3 2022100212 2007 46461 Mike Boone (13:12) M.Boone left tackle to DEN 44 for 7 yards (J.Abram; D.Deablo). 3 2 10 DEN LV DEN 37 13:12 19 16 nan nan nan 7 7 N 47 SINGLEBACK 6.000000 0.592704 0.620552 0.379448 -0.002451 0.002451 1.641006 -0.043580 nan nan nan nan 0.000000 2.000000 0.000000 2.000000
4 2022091900 1372 47857 Devin Singletary (8:33) D.Singletary right guard to TEN 32 for 3 yards (N.Jones; O.Adeniyi). TEN-O.Adeniyi was injured during the play. His return is Questionable. O.Adeniyi walks off. 2 1 10 BUF TEN TEN 35 8:33 7 7 nan nan nan 3 3 N 75 I_FORM 7.000000 0.470508 0.836290 0.163710 0.001053 -0.001053 3.686428 -0.167903 nan nan nan nan 0.000000 2.000000 0.000000 1.000000
In [62]:
values = {"tackles": -1.0, "assists": -1.0, "forced_fumbles": -1.0, "missed_tackles": -1.0}
play_tackles = play_tackles.fillna(value=values)

play_tackles = play_tackles.astype({"tackles":'int', "assists":'int', 
                                    "forced_fumbles":'int', "missed_tackles":'int',}) 

play_tackles.info()


ctt = pd.crosstab(play_tackles['offenseFormation'], play_tackles['tackles'])
ctt.columns = ['No tackles', 
                'Unsuccessful tackles',
               'Successful tackles',
               '2 successful tackles']

<class 'pandas.core.frame.DataFrame'>
Int64Index: 12486 entries, 0 to 12485
Data columns (total 39 columns):
 #   Column                            Non-Null Count  Dtype  
---  ------                            --------------  -----  
 0   gameId                            12486 non-null  int64  
 1   playId                            12486 non-null  int64  
 2   ballCarrierId                     12486 non-null  int64  
 3   ballCarrierDisplayName            12486 non-null  object 
 4   playDescription                   12486 non-null  object 
 5   quarter                           12486 non-null  int64  
 6   down                              12486 non-null  int64  
 7   yardsToGo                         12486 non-null  int64  
 8   possessionTeam                    12486 non-null  object 
 9   defensiveTeam                     12486 non-null  object 
 10  yardlineSide                      12319 non-null  object 
 11  yardlineNumber                    12486 non-null  int64  
 12  gameClock                         12486 non-null  object 
 13  preSnapHomeScore                  12486 non-null  int64  
 14  preSnapVisitorScore               12486 non-null  int64  
 15  passResult                        6105 non-null   object 
 16  passLength                        5634 non-null   float64
 17  penaltyYards                      615 non-null    float64
 18  prePenaltyPlayResult              12486 non-null  int64  
 19  playResult                        12486 non-null  int64  
 20  playNullifiedByPenalty            12486 non-null  object 
 21  absoluteYardlineNumber            12486 non-null  int64  
 22  offenseFormation                  12482 non-null  object 
 23  defendersInTheBox                 12481 non-null  float64
 24  passProbability                   12149 non-null  float64
 25  preSnapHomeTeamWinProbability     12486 non-null  float64
 26  preSnapVisitorTeamWinProbability  12486 non-null  float64
 27  homeTeamWinProbabilityAdded       12486 non-null  float64
 28  visitorTeamWinProbilityAdded      12486 non-null  float64
 29  expectedPoints                    12486 non-null  float64
 30  expectedPointsAdded               12485 non-null  float64
 31  foulName1                         592 non-null    object 
 32  foulName2                         25 non-null     object 
 33  foulNFLId1                        592 non-null    float64
 34  foulNFLId2                        25 non-null     float64
 35  tackles                           12486 non-null  int32  
 36  assists                           12486 non-null  int32  
 37  forced_fumbles                    12486 non-null  int32  
 38  missed_tackles                    12486 non-null  int32  
dtypes: float64(12), int32(4), int64(12), object(11)
memory usage: 3.6+ MB
In [63]:
# Heatmap
heatmap_trace = go.Heatmap(
    z=ctt,
    x=ctt.columns,
    y=ctt.index,
    colorscale='purples',
)

# Create figure and add trace
fig = go.Figure(heatmap_trace)

# Update layout
fig.update_layout(title='Heatmap of Successful/Unsuccessful tackles by Offense Formation')

# Display the figure
fig.show()
No tacklesUnsuccessful tacklesSuccessful tackles2 successful tacklesEMPTYI_FORMJUMBOPISTOLSHOTGUNSINGLEBACKWILDCAT
010002000300040005000Heatmap of Successful/Unsuccessful tackles by Offense Formation
plotly-logomark

We find that

  • the most of the successful tackles happened when the offensive team used either SHOTGUN or SINGLEBACK offense formations
  • the most of the missed tackle events happened when the offensive team used either SHOTGUN or SINGLEBACK offense formations
  • At the same time, we can see SHOTGUN and SINGLEBACK offense formations to be of the widest use in the plays listed in the dataset for the contest. So, the 'law of big numbers' takes the effect here.

Team Tackling Performance

In [64]:
# inspired by https://www.kaggle.com/code/sasakitetsuya/1st-step-data-summary-and-understanding

# Merge tackles data with plays data to get the defensive team for each play
team_tackles_df = tackles.merge(plays[['gameId', 'playId', 'defensiveTeam']], on=['gameId', 'playId'])

# Calculate total tackles and missed tackles for each team
team_performance = team_tackles_df.groupby('defensiveTeam').agg(
    total_tackles=('tackle', 'sum'),
    total_assists=('assist', 'sum'),
    missed_tackles=('pff_missedTackle', 'sum')
).reset_index()

team_performance['total_tackles_and_assists'] = team_performance['total_tackles'] + team_performance['total_assists']

# Sort teams by total tackles in descending order
team_performance_sorted = team_performance.sort_values(by='total_tackles_and_assists', ascending=False)
In [65]:
fig = go.Figure()

fig.add_bar(x=team_performance_sorted['defensiveTeam'],y=team_performance_sorted['total_tackles_and_assists'], 
            hovertext="Number of total tackles and assists", name="number_of_total_tackles_and_assists", 
            marker_color=px.colors.qualitative.Prism[0])
fig.add_bar(x=team_performance_sorted['defensiveTeam'],y=team_performance_sorted['total_tackles'], 
            hovertext="Number of total tackles", name="number_of_total_tackles", 
            marker_color=px.colors.qualitative.Prism[1])
fig.add_bar(x=team_performance_sorted['defensiveTeam'],y=team_performance_sorted['total_assists'], 
            hovertext="Number of assists in tackles", name="number_of_assists", 
            marker_color=px.colors.qualitative.Prism[2])
fig.add_bar(x=team_performance_sorted['defensiveTeam'],y=team_performance_sorted['missed_tackles'], 
            hovertext="Number of missed tackles", name="number_of_missed_tackles", 
            marker_color=px.colors.qualitative.Prism[3])

# Update visual layout
fig.update_layout(
    showlegend=True,
    width=800,
    height=400,
    autosize=False,
    margin=dict(t=30, b=0, l=5, r=5),
    template="plotly_white",
    yaxis_title="Number of tackles",
    title=dict(text='Total vs. Missing Tackles by Teams', font=dict(size=20), yref='paper')
)

# update font size at the axes
fig.update_coloraxes(colorbar_tickfont_size=10)
# Update font in the titles: Apparently subplot titles are annotations (Subplot font size is hardcoded to 16pt · Issue #985)
fig.update_annotations(font_size=12)
# Reduce opacity
fig.update_traces(opacity=0.75)
fig.show()
ATLCARMIAARINYJJAXSEAINDTBCHICINNELVDENGBWASPITDETNOBALDALLATENHOUMINKCBUFNYGLACPHICLESF0100200300400500600
number_of_total_tackles_and_assistsnumber_of_total_tacklesnumber_of_assistsnumber_of_missed_tacklesTotal vs. Missing Tackles by TeamsNumber of tackles
plotly-logomark

We can see that

  • The Atlanta Falcons (ATL) lead in total tackles+assists as well as in the total number of assists,
  • The Seattle Eagles (SEA) lead in total tackles
  • The Jacksonville Jaguars (JAX) lead in the number of missed tackles
  • In turn, the San Francisco 49ers (SF) have the fewest tackles among the teams.

However, it's essential to consider both total tackles, assists, and missed tackles when evaluating a team's defensive performance.

Win Rate by Teams¶

In [66]:
# inspired by https://www.kaggle.com/code/sasakitetsuya/1st-step-data-summary-and-understanding
# Calculate win counts for home and visitor teams
home_wins = games[games['homeFinalScore'] > games['visitorFinalScore']].groupby('homeTeamAbbr').size().reset_index(name='home_wins')
visitor_wins = games[games['visitorFinalScore'] > games['homeFinalScore']].groupby('visitorTeamAbbr').size().reset_index(name='visitor_wins')

# Merge win counts and calculate total wins for each team
team_wins = home_wins.merge(visitor_wins, left_on='homeTeamAbbr', right_on='visitorTeamAbbr', how='outer').fillna(0)
team_wins['total_wins'] = team_wins['home_wins'] + team_wins['visitor_wins']

# Calculate total games played by each team
total_games_played = games.groupby('homeTeamAbbr').size().reset_index(name='games_played')

# Calculate the total number of games played by each team
team_games_home = games['homeTeamAbbr'].value_counts()
team_games_visitor = games['visitorTeamAbbr'].value_counts()
total_games_per_team = team_games_home.add(team_games_visitor, fill_value=0)

# Merge win counts with total games played to calculate win rate
team_win_rate = team_wins.merge(total_games_played, left_on='homeTeamAbbr', right_on='homeTeamAbbr')

# Calculate win rate using the correct total games played
team_win_rate['games_played'] = team_win_rate['homeTeamAbbr'].map(total_games_per_team)
team_win_rate['win_rate'] = team_win_rate['total_wins'] / team_win_rate['games_played']

# Sort teams by corrected win rate in descending order
team_win_rate_sorted = team_win_rate[['homeTeamAbbr', 'win_rate']].sort_values(by='win_rate', ascending=True)


fig = px.bar(
    team_win_rate_sorted, 
    y='homeTeamAbbr', 
    x="win_rate", 
    orientation='h', 
    title='Winning Rate by teams', 
    height=900, 
    width=800,
    color_discrete_sequence=px.colors.qualitative.Prism,
)
fig.show()
00.20.40.60.81CARPITLVDETARINOCHIJAXINDGBLACLEDENWASTBATLSFNECINLACTENMIABALNYJSEANYGKCBUFDALMINPHI
Winning Rate by teamswin_ratehomeTeamAbbr
plotly-logomark

As we can see, the Philadelphia Eagles (PHI) have the highest win rate of 100%, indicating they won all their games in the data provided. In contrast, the Carolina Panthers (CAR) have the lowest win rate of 22.2%.

Win Rate and Tackling Performance of Teams

As we review the charts above, we can draw the intutive assumption on the lack of strong linear relations between the tackling performance of the teams and their win rates in the games.

The Philadelphia Eagles (PHI) who hold the highest win rate (1.0) in the season do not demonstrate outstanding tackling performnace (they are in the bottom 5 list of the teams in terms of the tackling performance metrics) In turn, the leaders of the tackling perfomrnace rating list do not demonstrate the high win rate (the win rate of Atlanta Falcons (ATL) is 0.4444, and the win rate of the Seattle Eagles (SEA) is 0.6667) Let's prove this intelligence by a bit of statistical analysis and visualizations. Let's look at the resepctive correlations.

In [67]:
dd = team_performance_sorted.merge(team_win_rate[['homeTeamAbbr', 'win_rate']], 
                                   left_on='defensiveTeam', 
                                   right_on='homeTeamAbbr')

d = dd[['total_tackles', 'total_assists', 'missed_tackles', 'total_tackles_and_assists', 'win_rate']]

# Compute the correlation matrix for numeric features
corr = d.corr()

# Generate a mask for the upper triangle
mask = np.triu(np.ones_like(corr, dtype=bool))

# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(11, 9))

# Generate a custom diverging colormap
cmap = sns.diverging_palette(230, 20, as_cmap=True)

# Draw the heatmap with the mask and correct aspect ratio
sns.heatmap(corr, mask=mask, cmap=cmap, vmax=.3, center=0, annot=True,
            square=True, linewidths=.5, cbar_kws={"shrink": .5})
Out[67]:
<Axes: >

We can see that

  • the missed_tackles does not have any strong- or medium-level correlations with win_rate
  • there is a medium-level negative correlation win_rate and total_assists performed by the team players; it is somewhat counter-intuitive at a glance, and it will require deeper analysis

CONCLUSION

Wise people say, there is yet-to-be-discovered rationale/intelligence behind most of the intuitive anticipations. Reviewing the body of knowledge on tackling success factors for Union Rugby substantiates there are good approaches identified by the research there.

The purpose of this essay is to scrutinize the previous findings as for the tackling range and player-in-tackle evaluation in Union Rugby. Plot the NFL data-driven insights to prove which metrics identified for tackling evaluation in Union Rugby are good for NFL tackle evaluation as well.

The current research can be extended for sure.